A non-optically active water quality physicochemical parameter remote sensing estimation method

Abstract

The application relates to a non-optically active water quality physicochemical parameter remote sensing estimation method, satellite remote sensing data is periodically inquired, retrieved and downloaded through a satellite remote sensing data service interface; a multi-dimensional feature set is constructed based on sampling time information, spatial position information and multi-band spectral information of the satellite remote sensing data; the multi-dimensional feature set is used as model input, a machine learning model is trained in combination with measured values of non-optically active water quality physicochemical parameters, and remote sensing estimation models of multiple non-optically active water quality physicochemical parameters are established; new measured values of water quality physicochemical parameters are acquired, when the new measured values meet a model updating trigger condition, controlled hot start updating is performed with parameters of a current remote sensing estimation model as initial parameters, and the current remote sensing estimation model is corrected. The method can realize full-link remote sensing monitoring from automatic data acquisition, multi-dimensional feature collaborative modeling to model self-optimization, and high-precision water quality physicochemical parameter remote sensing dynamic monitoring management is realized.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing and environmental monitoring technology, specifically relating to a remote sensing estimation method for non-optically active water quality physicochemical parameters. Background Technology

[0002] Lakes, as an important component of terrestrial aquatic ecosystems, undertake multiple ecological functions, including maintaining regional hydrological cycles, regulating climate, and protecting biodiversity. They are not only vital sources of water resources but also play an irreplaceable role in ecological balance and providing habitats for organisms. Lakes significantly influence water distribution and quality within their basins by regulating hydrological cycles, while simultaneously providing habitats for diverse aquatic species, thus maintaining the stability and health of the ecosystem. However, in recent years, intensified global climate change and increased human activity have led to a series of severe ecological challenges for lakes, particularly eutrophication and water quality deterioration. Eutrophication causes abnormally high concentrations of nutrients such as nitrogen and phosphorus in water bodies, inducing ecological disasters such as algal blooms, seriously affecting the aquatic ecological environment and water quality safety, and threatening the sustainable use of water resources.

[0003] Furthermore, water pollution and ecological degradation not only alter lake water quality but also pose a serious threat to the safety of the surrounding water environment. The continuous accumulation of pollutants, the decline of ecosystems, and the deterioration of water quality not only change the original ecological functions of lakes but may also trigger changes in regional hydrological cycles and climate patterns, thereby adversely affecting local agricultural production, the ecological environment, and the quality of human life. Therefore, conducting comprehensive monitoring of the physicochemical parameters of lake water quality and improving the ability to conduct refined assessments of surface water quality are particularly important. This will not only help strengthen water environment security but also provide strong support for the scientific management and sustainable utilization of water resources.

[0004] Traditional water quality monitoring methods have certain limitations, mainly reflected in low sampling point density, insufficient dynamic monitoring frequency, and high monitoring costs. Due to the limited spatial distribution of sampling points, traditional monitoring methods struggle to achieve comprehensive coverage of large-scale lake groups, especially in remote areas or vast lake clusters. Existing sampling data cannot reflect water quality changes in a timely manner, resulting in delayed monitoring results and affecting the timeliness of water quality management and assessment. Furthermore, traditional monitoring methods typically rely on manual sampling and laboratory analysis, which incurs high economic costs and time consumption, making it impossible to monitor dynamic changes in water quality parameters in real time.

[0005] In contrast, satellite remote sensing technology, with its high spatial coverage, periodic observation capabilities, and relatively low economic cost, has become an ideal choice for solving the aforementioned problems. Satellite remote sensing can achieve efficient water quality monitoring over large areas, without being limited by geographical location or environmental conditions, making it particularly suitable for long-term monitoring and dynamic assessment of water bodies such as lakes. Furthermore, by combining spatiotemporal spectral information, satellite remote sensing can provide multi-dimensional and refined water quality assessment results through multi-temporal and multi-band data sources. This provides strong technical support for the scientific monitoring and management of lake water quality.

[0006] Traditional remote sensing inversion algorithms for water quality parameters demonstrate high applicability when processing optically active parameters such as chlorophyll a. Changes in the concentration of these parameters typically induce changes in the remote sensing reflectance of characteristic spectral bands. For example, chlorophyll a has a reflectance peak at 705 nm and an absorption peak at 665 nm; these changes in characteristic spectral bands allow for the inversion of these parameters. However, for non-optically active parameters such as chemical oxygen demand (COD) and total phosphorus, traditional methods struggle to capture their spatial distribution because changes in their concentration do not significantly alter the spectral characteristics of remote sensing images. Summary of the Invention

[0007] The purpose of this invention is to provide a remote sensing estimation method for non-optically active water physicochemical parameters. This method utilizes satellite remote sensing data service interfaces to periodically acquire data, and combines multi-feature parameter machine learning modeling that considers spatiotemporal information with hot-start update to construct an estimation model. This method can achieve continuous, real-time, and accurate monitoring of the concentration changes of non-optically active water physicochemical parameters, improve the ability to monitor surface water quality, and strengthen water safety assurance and human monitoring management.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A remote sensing method for estimating non-optically reactive water physicochemical parameters, the method comprising:

[0010] Periodically query, retrieve, and download satellite remote sensing data through the satellite remote sensing data service interface;

[0011] A multidimensional feature set is constructed based on the sampling time information, spatial location information and multi-band spectral information of satellite remote sensing data;

[0012] Using the multidimensional feature set as model input, and combining the measured values ​​of non-optically active water quality physicochemical parameters, a machine learning model is trained to establish multiple remote sensing estimation models for non-optically active water quality physicochemical parameters.

[0013] New measured values ​​of water quality physicochemical parameters are acquired. When the new measured values ​​meet the model update triggering conditions, a hot start update is performed using the parameters of the current remote sensing estimation model as the initial parameters to correct the current remote sensing estimation model. The triggering condition is that the model's verification error exceeds a preset threshold.

[0014] In some embodiments of the present invention, the periodic querying, retrieval, and downloading of satellite remote sensing data includes:

[0015] A search task is triggered every preset time interval;

[0016] The retrieval task is based on product type, sensor / platform, track number, cloud cover threshold, and AOI construction interface query parameters;

[0017] Retrieve search results by calling the interface through the communication protocol;

[0018] For each item in the search results, its product ID and download link are parsed, and the data is automatically downloaded and stored in the database.

[0019] In some embodiments of the present invention, the time information is constructed using feature parameters constructed by the following formula:

[0020]

[0021] In the formula, DayOfYear is the date ordinal number of the sampling date in the corresponding year.

[0022] In some embodiments of the present invention, the spatial location information includes the longitude and latitude information corresponding to the measured sampling location.

[0023] In some embodiments of the present invention, the spectral information includes reflectance of different bands of satellite remote sensing data and combinations of reflectance of different bands, including:

[0024] R rs (blue) R rs (Green) R rs (red) R rs (red edge) R rs (narrow NIR);

[0025] R rs (blue) / R rs (Green) R rs (red) / Rrs (Green) R rs (red edge) / R rs (Green) R rs (rededge) / R rs (red);

[0026] in R rs (blue) R rs (Green) R rs (red) R rs (red edge) R rs (narrow NIR) refers to the remote sensing reflectance of blue light, green light, red light, red edge, and narrow near-infrared bands, respectively.

[0027] In some embodiments of the present invention, the machine learning model is an extreme gradient boosting tree.

[0028] In some embodiments of the present invention, when new measured values ​​of water quality physicochemical parameters are obtained, the new measured values ​​are added to historical training data to form an updated training dataset, and a hot start update is performed with the current estimation model as the initial parameter to obtain a candidate updated model.

[0029] In some embodiments of the present invention, the correction of the remote sensing estimation model includes:

[0030] The candidate update model is validated using the newly added measured values, and the validation error is calculated.

[0031] When the verification error is less than or equal to the preset error threshold, the candidate update model is retained until the next newly added measured value arrives.

[0032] When the verification error is greater than the preset error threshold, the time window reduction mechanism is triggered. Only the historical data within the most recent period or the most recent time span is selected and the newly added measured value is used to form a reduced and updated training dataset. The current estimated model parameters are used as the initial parameters to perform a hot start update. This process is repeated until the verification error is less than or equal to the preset error threshold.

[0033] In some embodiments of the present invention, the non-optically active water physicochemical parameters are water quality parameters without stable spectral response characteristics, including chemical oxygen demand, permanganate index, ammonia nitrogen, dissolved oxygen, total phosphorus, and total nitrogen.

[0034] In some embodiments of the present invention, the data source of the satellite remote sensing data is an MSI satellite sensor.

[0035] This invention first automates the scheduling and near-real-time acquisition of satellite remote sensing image data based on a data interaction interface. Then, it integrates temporal features, spatial location features, and multi-band spectral reflectance features to construct a multi-dimensional feature set. An extreme gradient enhancement tree algorithm is then used to establish a remote sensing estimation model for various non-optically active water quality parameters such as chemical oxygen demand (COD), ammonia nitrogen, and total phosphorus. Furthermore, an adaptive hot-start update mechanism is designed. When new measured water quality parameter data arrives, a controlled hot-start update is performed using the previous inversion model as the initial parameters to obtain a candidate updated model. Based on the verification error, it is determined whether to directly apply the updated model or to rebuild the training set by reducing the time window and perform another hot-start update to obtain the final updated model. This achieves a complete technical process from automatic remote sensing data acquisition and multi-dimensional feature fusion modeling to dynamic model updating.

[0036] The method of this invention can realize the entire remote sensing monitoring chain from automatic data acquisition and multi-dimensional feature collaborative modeling to autonomous model optimization. It enables continuous, real-time and accurate monitoring of the concentration changes of non-optically active water physicochemical parameters, providing a solid technical foundation for high-precision remote sensing dynamic monitoring and management of water physicochemical parameters, and helping to strengthen water safety and sustainable management.

[0037] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below may be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other. Furthermore, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.

[0038] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description

[0039] The accompanying drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in the various figures may be denoted by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the invention will now be described by way of example and with reference to the accompanying drawings, wherein:

[0040] Figure 1 This is a flowchart of the method of the present invention.

[0041] Figure 2 This refers to the model accuracy of the GWRF algorithm in the inversion of water quality physicochemical parameters.

[0042] Figure 3 This refers to the model accuracy of the RFR algorithm in the inversion of water quality physicochemical parameters.

[0043] Figure 4 This refers to the model accuracy of the SVR algorithm in the inversion of water quality physicochemical parameters.

[0044] Figure 5 This refers to the model accuracy of the XGBoost algorithm in the inversion of water quality physicochemical parameters.

[0045] Figure 6 It is the accuracy of the water quality physicochemical parameter inversion model under the first feature input (spectral information).

[0046] Figure 7 It is the accuracy of the water quality physicochemical parameter inversion model under the second feature input (spectral and spatial information).

[0047] Figure 8 It is the accuracy of the water quality physicochemical parameter inversion model under the third feature input (spectral and temporal information).

[0048] Figure 9 It is the accuracy of the water quality physicochemical parameter inversion model under the fourth feature input (spectral, temporal and spatial information).

[0049] Figure 10 It is a diagram showing the accuracy of a remote sensing estimation model for water quality physicochemical parameters under a certain time function.

[0050] Figure 11 This is an accuracy diagram of the remote sensing estimation model for water quality physicochemical parameters under the time function of this invention.

[0051] Figure 12 This is a comparison chart of the model's accuracy before and after optimization under the adaptive hot start update mode; where (A) is the validation result of the model trained based on historical and current month (May) measured data in May; and (B) is the validation result of the model trained based on measured data from the past three months (March-May) in May.

[0052] Figure 13 This is a spatial distribution map of water quality assessment for a lake on September 23, 2024 (taking dissolved oxygen DO as an example). Detailed Implementation

[0053] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0054] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, in which numerous illustrative embodiments are shown. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, as the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0055] Example 1

[0056] This embodiment uses a lake as an example to further describe the technical solution of the present invention.

[0057] This invention relates to a remote sensing estimation method for surface water physicochemical parameters that takes into account spatiotemporal information. Based on a satellite remote sensing data service interface conforming to the OpenSearch API protocol, an automated scheduling mechanism is constructed to achieve near real-time acquisition of remote sensing image data. A multi-dimensional feature set is constructed by fusing temporal features, spatial location features, and multi-band spectral reflectance features. An extreme gradient enhancement tree algorithm is used to establish a remote sensing estimation model for non-optically active water quality parameters such as chemical oxygen demand (COD), ammonia nitrogen, and total phosphorus. Simultaneously, an adaptive hot-start update mechanism is designed. When new measured water quality parameter data arrives, a controlled hot-start update is performed using the previous inversion model as the initial parameters to obtain a candidate updated model. If the verification error does not exceed a preset threshold, the updated model is directly adopted; otherwise, it is determined that the model input-output relationship has drifted. The training set is reconstructed by reducing the time window, and a hot-start update is performed again to obtain the final updated model. This invention enables end-to-end water quality remote sensing monitoring from automatic data acquisition and feature fusion modeling to autonomous model updating, providing technical support for water environment management and protection.

[0058] As an exemplary description, the implementation of the aforementioned method will be specifically explained below with reference to the accompanying drawings.

[0059] The method for remote sensing estimation of surface water physicochemical parameters based on spatiotemporal information of the present invention includes three parts: automatic acquisition of remote sensing data, construction and optimization of water quality physicochemical parameter models, and production of standardized remote sensing products. The core process is as follows: Figure 1 As shown, the specific steps include the following:

[0060] 1) Based on the satellite remote sensing data service interface conforming to the OpenSearch API protocol, a timed task scheduling mechanism is constructed to periodically and automatically call the satellite remote sensing data service interface. Within a 4-hour time window after the satellite remote sensing images are distributed, the automatic query, retrieval and near real-time acquisition of remote sensing data are realized.

[0061] The scheduled task mechanism includes: triggering a retrieval task every 4 hours by a Python timer; constructing OpenSearch query parameters based on product type, sensor / platform, orbit number, cloud cover threshold, and AOI, and obtaining retrieval results by calling the OpenSearch interface via HTTP(S); parsing the product ID and download URL of the entries in the retrieval results to realize the timed automated retrieval, download, and storage of satellite remote sensing data.

[0062] Following the standard lake environment remote sensing processing workflow, lake remote sensing data preprocessing and quality control were performed: the DSF algorithm integrated into the ACOLITE software was used to perform atmospheric correction on the MSI data to obtain the water remote sensing reflectance (…). R rs Data; land and water are separated using the bimodal thresholding method, and the land is masked; pixels with atmospheric top reflectance greater than 0.3 in any band, or atmospheric top reflectance greater than 0.005 in the 1373 nm band, are considered cloud pixels and masked; aquatic vegetation products are used to mask aquatic vegetation areas.

[0063] 2) Based on the sampling information and the MSI images after preprocessing and quality control, a multi-dimensional feature set is constructed as the input variable of the machine learning model. It is divided into three parts: sampling time information, spatial location information and spectral information.

[0064] ① The time information is represented by a sampling time periodic function constructed using the following formula:

[0065] (1)

[0066] In the formula, DayOfYear is the date ordinal number of the sampling date in the corresponding year.

[0067] ② Spatial location information: Longitude and latitude information corresponding to the measured sampling locations.

[0068] ③ Spectral information includes remote sensing reflectance in 5 bands: R rs (492) R rs (560) R rs (665) R rs (705) R rs (865); Ratio of remote sensing reflectance in 4 bands: R rs (492) / Rrs(560), R rs (665) / Rrs(560),R rs (705) / Rrs(560), R rs (705) / R rs (665).

[0069] 3) Based on the information obtained in 2), the machine learning model is trained using the measured water quality parameter data of non-optically active water physicochemical parameters at the sampling points as ground truth, and multiple remote sensing estimation models for non-optically active water physicochemical parameters are established (first-round initial model). Taking the XGBoost model as an example, K-fold cross-validation is used during model tuning.

[0070] 4) The estimation model (inversion model) is continuously corrected based on the adaptive hot start update mechanism.

[0071] Based on the constructed remote sensing estimation model, the corresponding non-optically active water quality physicochemical parameters are estimated in near real-time using remote sensing.

[0072] When newly measured water quality physicochemical parameter data arrives, it is determined whether the preset model update trigger condition is met (the model's validation error exceeds a preset threshold). If the trigger condition is met, the current model parameters (the parameters of the initial model in the first round) are used as the initial constraint state, and a hot start update of the model parameters is performed to maintain the model estimation accuracy and stability during long-term online operation. Specifically:

[0073] When new measured water quality parameters arrive, the model is iteratively updated. The new measured water quality parameters are merged with historical training data to form a full update dataset. The parameters of the current estimated model are used as initial parameters for a controlled hot-start update to obtain candidate models. Subsequently, the candidate models are validated using the new data. If the validation error is less than or equal to a preset threshold, the current water quality response relationship is considered stable, and the candidate model is retained as the final inversion model. If the error exceeds the threshold, the input-output relationship is considered to have undergone a trend drift over time, triggering a time window reduction mechanism. Old historical data is removed, and only recent window data and new data are selected to reconstruct a reduced training set and perform a hot-start update, thereby obtaining a final model with dynamic adaptability, improving adaptability to dynamic changes in the system and inversion accuracy.

[0074] Example 2

[0075] This embodiment, based on Embodiment 1, compares the impact of different solutions on the results.

[0076] (1) This step compared different machine learning algorithms, including Geographically Weighted Random Forest (GWRF), Random Forest Regression (RFR), Supported Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost). The scatter plot accuracy of each model is as follows: Figures 2-5 As shown, the optimal machine learning algorithm, namely the extreme gradient boosting tree, was selected through comparison.

[0077] (2) This step systematically evaluated the impact of different input features on model accuracy (e.g., Figures 6-9 The results showed that introducing time and space functions effectively improved model performance. Using RMSE as the evaluation metric, the improvement ranged from 7.69% to 31.85%.

[0078] (3) This step provides a comparison of different time functions, and the influence of the input parameters of the two different time functions on the accuracy of the water quality physicochemical parameters retrieved by the model is as follows: Figure 10 ( ), Figure 11 As shown, the time function of the present invention has higher estimation accuracy while retaining more samples. Taking total nitrogen estimation as an example, the model constructed in this embodiment increases the effective data volume by 30.62% and improves the accuracy by 14.29% (taking RMSE as an example), which greatly improves the accuracy of describing daily water quality changes.

[0079] (4) This step compares the effects of different hot start update methods on the results.

[0080] As training data accumulates, the model dynamically adjusts its accuracy through a warm start. Because water physicochemical parameters are influenced by human intervention or natural changes, directly updating the model when newly measured data differs significantly from the historical training set in terms of temporal or spatial distribution may lead to a decrease in accuracy. Figure 12 As shown. At this point, the model will select the current month as the starting point to construct a time window of nearly three months in subsequent updates to ensure the accuracy and stability of the model's estimation in terms of temporal and spatial distribution. Figure 13 The spatial distribution of various water quality parameters of a lake estimated by this method on September 23, 2024 (taking DO as an example) is further shown, demonstrating the output capability of the model in practical regional applications.

[0081] The above method enables the establishment of a remote sensing estimation method for surface water quality physicochemical parameters that takes into account spatiotemporal information, achieving high monitoring accuracy. Employing the method of this invention helps improve the ability to conduct refined assessments of surface water quality, and strengthens water body safety assurance and human monitoring management.

[0082] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A remote sensing method for estimating non-optically active water quality physicochemical parameters, characterized in that, The method includes: Periodically query, retrieve, and download satellite remote sensing data through the satellite remote sensing data service interface; A multidimensional feature set is constructed based on the sampling time information, spatial location information and multi-band spectral information of satellite remote sensing data; Using the multidimensional feature set as model input, and combining the measured values ​​of non-optically active water quality physicochemical parameters, an extreme gradient enhancement tree is trained to establish multiple remote sensing estimation models for non-optically active water quality physicochemical parameters. When new measured values ​​of water quality physicochemical parameters are obtained, the new measured values ​​are added to the historical training data to form an updated training dataset. The current remote sensing estimation model is used as the initial parameter to perform a hot start update to obtain a candidate update model. The candidate update model is then verified using the new measured values ​​and the verification error is calculated. When the verification error is less than or equal to the preset error threshold, the candidate update model is retained until the next newly added measured value arrives. When the verification error is greater than the preset error threshold, the time window reduction mechanism is triggered. Only the historical data within the most recent period or the most recent time span is selected and the newly added measured value is used to form a reduced and updated training dataset. The current estimated model parameters are used as the initial parameters to perform a hot start update. This process is repeated until the verification error is less than or equal to the preset error threshold.

2. The method according to claim 1, characterized in that, The periodic querying, retrieval, and downloading of satellite remote sensing data includes: A search task is triggered every preset time interval; The retrieval task is based on product type, sensor / platform, track number, cloud cover threshold, and AOI construction interface query parameters; Retrieve search results by calling the interface through the communication protocol; For each item in the search results, its product ID and download link are parsed, and the data is automatically downloaded and stored in the database.

3. The method according to claim 1, characterized in that, The time information is constructed using the following feature parameters: ； In the formula, DayOfYear is the date ordinal number of the sampling date in the corresponding year.

4. The method according to claim 1, characterized in that, The spatial location information includes the longitude and latitude information corresponding to the measured sampling location.

5. The method according to claim 1, characterized in that, The spectral information includes reflectance in different bands of satellite remote sensing data and combinations of reflectance in different bands, including: R rs (blue)、R rs (Green)、R rs (red)、R rs (red edge)、R rs (narrow NIR); R rs (blue) / R rs (Green)、R rs (red) / R rs (Green)、R rs (red edge) / R rs (Green)、R rs (rededge) / R rs (red); Where R rs (blue), R rs (Green), R rs (red), R rs (red edge), R rs (narrow NIR) refers to the remote sensing reflectance of blue light, green light, red light, red edge, and narrow near-infrared bands, respectively.

6. The method according to claim 1, characterized in that, The non-optically active water physicochemical parameters are water quality parameters without stable spectral response characteristics, including chemical oxygen demand, permanganate index, ammonia nitrogen, dissolved oxygen, total phosphorus, and total nitrogen.

7. The method according to claim 1, characterized in that, The data source for the satellite remote sensing data is the MSI satellite sensor.

Images